An estimate is a forecast, so give it a range
Single-number estimates are a promise nobody can keep. Ranges are honest and, oddly, easier to sell.
Somebody asks how long a feature will take and you say six weeks. You have just made a commitment with a hidden probability attached, and the person hearing it assumed that probability was close to one hundred per cent. It was more like fifty.
We give ranges instead, along with the thing that would narrow them.
Five to nine weeks. It is at the low end if the payment provider's sandbox behaves like their docs, and at the high end if we have to reverse-engineer their webhook ordering. We will know within four days.
Why clients prefer this
The objection we expected was that ranges look evasive. In practice the opposite happens. A range with a named cause attached is legible: it tells a client what the uncertainty is about and when it resolves. A single number tells them nothing except what to be angry about later.
It also changes the conversation from negotiation to investigation. Instead of arguing about whether six weeks could be four, we agree to spend four days finding out which end of the range we are on.
Three rules we hold to
- Every range names the specific unknown driving its width
- Every range names the date the unknown resolves
- When a range moves, the client hears it that week, with the reason
The third one is the one that takes discipline. A slip you report in week two is a scheduling conversation. The same slip reported in week ten is a dispute about your competence.